OpenAI Launches GPT-5.5 With Major Gains in Agentic Coding and Computer Use

OpenAI has introduced GPT-5.5, its latest model focused on coding, computer use, knowledge work, and scientific research. The company says the new model can take on messy, multi-part tasks with more autonomy while matching GPT-5.4 per-token latency in real-world serving and using fewer tokens on the same Codex tasks.



OpenAI says GPT-5.5 is its strongest agentic coding model so far. On Terminal-Bench 2.0, which measures complex command-line workflows, the model scored 82.7%, up from 75.1% for GPT-5.4. On SWE-Bench Pro, which evaluates real-world GitHub issue resolution, it reached 58.6%, and OpenAI says it solved more tasks end-to-end in a single pass than earlier versions. The company also says early testing showed stronger performance on long-horizon coding tasks, ambiguous failures, and carrying changes through a larger codebase.


The update also pushes further into general computer-based work. OpenAI says GPT-5.5 is better than GPT-5.4 in Codex at generating documents, spreadsheets, and slide presentations, and that its computer-use abilities make it better at moving across tools, checking results, and navigating interfaces. Internally, OpenAI says its finance team used Codex with GPT-5.5 to review 24,771 K-1 tax forms spanning 71,637 pages, helping accelerate the process by two weeks compared to the prior year.



On research tasks, OpenAI says GPT-5.5 improved over GPT-5.4 on GeneBench and reached 80.5% on BixBench, which it describes as leading performance among models with published scores. The company also says an internal version of GPT-5.5 with a custom harness helped discover a new proof involving off-diagonal Ramsey numbers, later verified in Lean.

OpenAI credits part of the performance gain to infrastructure work built around NVIDIA systems. The company says GPT-5.5 was co-designed for, trained with, and served on NVIDIA GB200 and GB300 NVL72 systems. It also says Codex analyzed weeks of production traffic patterns and helped produce new load-balancing and partitioning heuristics that increased token generation speeds by more than 20%.



The release arrives with tighter safeguards as OpenAI expands controls around advanced cyber and biology-related misuse. The company says GPT-5.5's biological/chemical and cybersecurity capabilities are treated as "High" under its Preparedness Framework, while noting the model did not reach the "Critical" cybersecurity level. OpenAI is also expanding access through its Trusted Access for Cyber program for verified defenders.


GPT-5.5 is rolling out today to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex. GPT-5.5 Pro is rolling out to Pro, Business, and Enterprise users in ChatGPT, with both models set to reach the API soon. In Codex, GPT-5.5 is also available on Edu and Go plans with a 400K context window, and in a Fast mode that generates tokens 1.5x faster at higher cost.